Exploring Variation and Predictors of Residential

Int. J. Environ. Res. Public Health 2010, 7, 3211-3224; doi:10.3390/ijerph7083211 OPEN ACCESS

International Journal of Environmental Research and Public Health ISSN 1660-4601 www.mdpi.com/journal/ijerph Article

Exploring Variation and Predictors of Residential Fine Particulate Matter Infiltration Nina A. Clark 1, Ryan W. Allen 2, Perry Hystad 3, Lance Wallace 4, Sharon D. Dell 5, Richard Foty 5, Ewa Dabek-Zlotorzynska 6, Greg Evans 7 and Amanda J. Wheeler 1,* 1

2

3

4 5

6

7

Health Canada, 269 Laurier Ave West, Ottawa, Ontario K1A 0K9, Canada; E-Mail: [email protected] (N.A.C.) Simon Fraser University, 8888 University Drive, Burnaby, British Columbia V5A 1S6, Canada; E-Mail: [email protected] (R.W.A.) University of British Columbia, 2206 East Mall, Vancouver, British Columbia V6T 1Z3, Canada; E-Mail: [email protected] (P.H.) 11568 Woodhollow Ct, Reston, VA 20191, USA; E-Mail: [email protected] (L.W.) The Hospital for Sick Children, 555 University Avenue, Toronto, Ontario M5G 1X8, Canada; E-Mail: [email protected] (S.D.D.); [email protected] (R.F.) Environment Canada, 335 River Road, Ottawa, Ontario K1V 1C7, Canada; E-Mail: [email protected] (E.D.-Z.) University of Toronto, 200 College Street, Toronto, Ontario M5S 3E5, Canada; E-Mail: [email protected] (G.E)

* Author to whom correspondence should be addressed; E-Mail: [email protected]; Tel.: +1-613-948-3686; Fax: +1-613-948-8482. Received: 30 June 2010; in revised form: 12 August 2010 / Accepted: 13 August 2010 / Published: 16 August 2010

Abstract: Although individuals spend the majority of their time indoors, most epidemiological studies estimate personal air pollution exposures based on outdoor levels. This almost certainly results in exposure misclassification as pollutant infiltration varies between homes. However, it is often not possible to collect detailed measures of infiltration for individual homes in large-scale epidemiological studies and thus there is currently a need to develop models that can be used to predict these values. To address this need, we examined infiltration of fine particulate matter (PM2.5) and identified determinants of infiltration for 46 residential homes in Toronto, Canada. Infiltration was

Int. J. Environ. Res. Public Health 2010, 7

3212

estimated using the indoor/outdoor sulphur ratio and information on hypothesized predictors of infiltration were collected using questionnaires and publicly available databases. Multiple linear regression was used to develop the models. Mean infiltration was 0.52 ± 0.21 with no significant difference across heating and non-heating seasons. Predictors of infiltration were air exchange, presence of central air conditioning, and forced air heating. These variables accounted for 38% of the variability in infiltration. Without air exchange, the model accounted for 26% of the variability. Effective modelling of infiltration in individual homes remains difficult, although key variables such as use of central air conditioning show potential as an easily attainable indicator of infiltration. Keywords: air exchange; air quality; indoor; infiltration; fine particulate matter; PM2.5; residential; sulphur

1. Introduction Exposure to outdoor particulate matter (PM) has been linked to a wide variety of health effects [1-3]. However, exposure assessment in epidemiological studies remains a challenge and misclassification limits the power of many studies to find true effects [4,5]. A nearly universal source of exposure misclassification in studies of outdoor air pollution results from estimating exposure based on outdoor levels alone. Indeed, individuals‘ true exposures are determined by their time-activity patterns and levels of pollutants in each microenvironment where they spend time—typically indoors at home [6,7]. Surveys of activity patterns for Canada and the United States suggest that adults spend an average of about 88% of their time indoors, 66% of which is in their homes [8-10]. Exposure misclassification occurs because infiltration of outdoor pollution can vary significantly between residences and also within residences across time [11-13]. There are important differences between indoor and outdoor PM composition in addition to differences in concentration. Outdoor generated PM contains relatively more sulphates, nitrates, strong acids, and toxic metals as compared with PM generated in homes, which contains more house dust, endotoxins, mold spores, and fresh combustion products [9]. Consequently, indoor and outdoor generated PM appear to have different health effects. Health effects of indoor PM2.5 have not been thoroughly studied, but have been linked to respiratory problems [14-16]. Several studies that measured exposure to both indoor and outdoor PM2.5 have found that outdoor-generated PM generally showed stronger relationships with respiratory [17-19] and cardiovascular [18] markers than total or indoor-generated PM2.5 exposure. Therefore, differences in infiltration of outdoor generated PM2.5 across homes could introduce exposure misclassification and may bias health effect estimates [20]. Infiltration differences across homes have the potential to cause differential misclassification, where the degree of misclassification depends on other variables in the analysis (such as disease status). For example, infiltration has been found to depend on characteristics that may also influence disease status, such as socioeconomic status [21,22]. Unlike non-differential misclassification, which generally


3213

causes a bias to the null, differential misclassification may cause unexpected effects in health effect estimates by causing a bias toward or away from the null [20]. The infiltration factor (Finf) is defined as the equilibrium proportion of outdoor PM that penetrates indoors and remains suspended [23]. This is determined by the particle penetration efficiency (P, unitless), particle deposition rate (k, h−1), and the air exchange rate (a, h−1) according to the following equation: Pa Finf  (1) ak Various studies have estimated Finf for residential homes. Most commonly they have employed sulphur or sulphate as a tracer of outdoor PM2.5 [24-28], although indoor and outdoor measures of PM2.5 have also commonly been used to estimate Finf [11,13,21,29-31]. Sulphur has few indoor sources and has been demonstrated to be representative of total outdoor infiltrated PM2.5 [9,32,33]. Though a number of studies have estimated Finf in individual homes and some panel studies have considered Finf in epidemiologic analyses [17-19], the need to include both indoor and outdoor pollution measurements to estimate Finf has prevented inclusion of infiltration in large-scale epidemiologic studies. The inclusion of infiltration ‗modification‘ factors has been limited to air conditioning variables, such as community-wide prevalence of central air conditioning [34,35]. Given the importance of Finf (studies have shown Finf differences can modify indoor exposure rates by a magnitude of 2–4 [11,21,24]), it is important to further examine how Finf varies between homes and what drives these differences. The goal of this study was to determine household factors that influence Finf, as indicated by the sulphur tracer method [27] for homes in Toronto, Ontario. 2. Experimental Section 2.1. Data Collection Sixty homes from the Toronto area were recruited for the study. Homes were randomly selected from among 1,500 owner-occupied homes with secure backyards participating in the population-based Toronto Child Health Evaluation Questionnaire (T-CHEQ) study [36]. The T-CHEQ sample as a whole (n = 5,619) is representative of the population in terms of socioeconomic and housing variability as compared to the 2001 Canada census [36], however, the criteria of home-ownership with backyard biased the current study to a somewhat higher socioeconomic status. Fifty homes were initially recruited to participate in the indoor and outdoor sampling in 2006 (August to November) and an additional 10 homes were recruited to participate in 2007 (July). Participating households completed a baseline questionnaire before sampling began. The questionnaire was a detailed home assessment, including variables that were anticipated to influence Finf: home age and size [11,37], home value [21], heating and cooling systems used [21,24], presence of air filters [11,17], presence of storm windows [11,24], and number of residents and pets in the home (which have been observed to increase Finf due to increased air exchange) [37,38]. The assessment was completed by trained technicians along with the home owner during an inspection of the home. Unfortunately, it was not possible to collect information on activities in the home during sampling such as cooking, cleaning, window opening and air conditioner use.


3214

Each home was sampled over a five-day period, with approximately 4 homes sampled concurrently. Most inspections began in the middle of the week and included weekdays and weekend days. Indoor measurements were collected at breathing height inside participants‘ homes, typically in the family or living room where participants spent the majority of their time. When possible, measurements were collected away from ventilation ducts, fireplaces, and TVs as these may cause elevated concentrations of particles. Outdoor samplers were located in the backyard of the participants‘ homes several meters away from the house and away from any combustion sources such as barbeques, automobiles and other localized outdoor sources of air pollution. The outdoor sampling height was approximately 1.5 meters above ground. Indoor and outdoor PM2.5 mass was sampled over the entire 5-day period using a Chempass Multi-Component Sampling System (Model 3400, R&P/Thermo Scientific, Waltham, MA). The indoor and outdoor systems used a BGI pump with an AC adapter. The sampler target flow rate was 1.8 L/minute and flow rates more than 20% different from the target rate were considered invalid. Flow rates were assessed pre and post sampling using a soap bubble flow meter (AP Buck, Orlando, FL). The samples were collected on 37 mm TeflonTM filters, which were measured gravimetrically following US EPA quality assurance guidelines [39]. Briefly, filters were pre-conditioned for a minimum of 24 h before weighing at a constant air temperature of 21 oC (±0.5 oC) and constant relative humidity (RH) of 40% (±1%). All samples were analyzed for total sulphur by a Panalytical Epilson 5 energy dispersive x-ray fluorescence (ED-XRF) spectrometer with a 600-watt gadolinium target tube as an excitation source. Micromatter® standard reference samples were used for the quantitative calibration of the system. A calibration check was performed by analyzing NIST standards (SRM1832 and SRM1833). Following the measurement process, data treatment was performed using complex algorithms that corrected for background signal, spectral interferences, and X-ray signal loss due to absorption by the sample mass. Blank filters were used for background subtraction. Typical analytical limits of detection and uncertainty were estimated to be 3.0 ng/cm2 and 5%, respectively. The indoor temperature and relative humidity were measured using HOBO Temperature Relative Humidity Data Logger (U10-003, Onset Corporation, Bourne, MA, USA) at three-minute time intervals. Outdoor meteorological data were obtained from the National Climate Data and Information Archive of Environment Canada [40]. Air exchange rate (AER) (the rate at which outdoor air replaces indoor air) was measured in the 50 homes monitored in 2006 using a perfluorocarbon tracer (PFT) source emitting a tracer at a constant rate. This was captured by a capillary adsorption tube (CAT), a passive sampler containing a small amount of activated charcoal adsorbing material [41]. Four PFTs were placed in the corners of the floor where the indoor equipment was located, and one CAT was placed near the center of the floor at head height. After exposure, the CAT was shipped to the lab and analyzed by gas chromatography with an electron capture detector (GC/ECD). Air exchange rate was then calculated using the concentration of PFT detected on the CAT and the volume of the home (estimated by multiplying square footage of the home by the technician‘s estimate of ceiling height for the home). Air exchange was measured on the main living floor (in family or living room) only as this has been demonstrated to be a relatively accurate estimate of air exchange for the house: Wallace et al. [42] found inter-floor differences of less than 10% in a three-story home measured over a year. However, it was not possible


3215

to assess factors that may influence mixing of air in the home, such as closed doors. Closed doors to rooms or whole floors of the home for extended periods of time (e.g., nighttime) would reduce the effective volume of the home and thereby cause an underestimate of air exchange. In addition to information collected during exposure monitoring, other variables of interest were collected from publicly available tax records [43] and geographic information system (GIS) data. Specifically, information was obtained on home size and market value from the Municipal Property Assessment Corporation (MPAC) of Ontario, as these factors have been observed to influence Finf [21,37]. GIS data were used to determine the distance of each home to the nearest highway. Roadway classifications in the CanMap RouteLogistics dataset were used, including highways and expressways (400 series highways and expressways). 2.2. Statistical Methods Infiltration was calculated using the ratio of the indoor concentration and the outdoor concentration of sulphur (Equation 2).

Finf S 

Sin Sout

(2)

Linear regression was used to examine the association between housing and climatic variables and Finf. Right-skewed variables were log-normally transformed for regression modelling. Ambient temperature was examined as an effect modifier due to its influence on window opening behaviour, use of air conditioners and indoor/outdoor temperature differences. This was included as an interaction term indicating whether outdoor temperature was above or below 18 °C at time of monitoring. Many of the housing and climate characteristics are expected to be correlated and thus bivariate regressions may only indicate an indirect effect. However, variables without a direct effect may still be helpful for predicting Finf in studies with limited data, and are therefore presented along with multiple linear regression models. To understand the best predictors of Finf, variables that were associated with Finf (p < 0.10) in a simple linear regression were entered into multiple linear regression models. Two sets of models were examined: one set in which air exchange rate was a potential predictor variable and another set in which it was not, as this variable is often not available in larger-scale epidemiology studies. A stepwise procedure was used to select the most appropriate models, with p < 0.10 as the cut-off for inclusion of variables. The guideline of a minimum of 10 observations per degree of freedom was followed to guard against over-fitting. Collinearity between variables was examined using Pearson and Spearman correlation coefficients and t-tests where appropriate, as well as using the ―collin‖ option available for the REG procedure in SAS 9.1 (SAS Institute Inc., Cary, NC, USA). 3. Results Table 1 provides descriptive data for Finf, pollutant measurements, and home and climate characteristics that were hypothesized a priori to have an effect on Finf. Of the 60 participating homes, 53 were analyzed for sulphur content indoors and outdoors. Of the calculated Finf values, two homes had an indoor/outdoor sulphur ratio greater than one (2.89 and 3.09) and likely indicate an undetected


3216

indoor source of sulphur, for example, an unreported humidifier in the home [9]. Only values of Finf between 0 and 1 are physically valid; therefore, these two homes were excluded from further statistical analysis. An additional five homes had a sulphur ratio >0.90 (ranging from 0.93–0.98) without a correspondingly high air exchange rate; these homes were also excluded from further analysis, leaving a total of 46 homes for analysis. As compared with included homes, the excluded homes were more likely to have higher levels of indoor PM2.5 (13.9 ± 7.08 µg/m3 vs. 8.17 ± 5.18 µg/m3, p-value = 0.01) and indoor sulphur (1.08 ± 0.35 µg/m3 vs. 0.46 ± 0.31 µg/m3, p < 0.001) (consistent with having an indoor source). They did not differ with respect to other analyzed variables. Additional homes were excluded from specific analyses when covariate data was not available; this was most significant for air exchange which was not measured in the 10 homes sampled in 2007. The participating home-owners were of higher socioeconomic status as compared with the general Canadian population, but home values were representative of the Greater Toronto Area, where the average single-family home price is $435,064 [44]. Table 1. Baseline and measured characteristics of 46 homes in Toronto. For household characteristics that were not obtained via questionnaire, the data source is indicated in brackets. Measurement Characteristics Finf PM2.5 indoors PM2.5 outdoors Sulphur indoors Sulphur outdoors Air exchange Indoor temperature Indoor relative humidity Outdoor temperature Outdoor relative humidity Household Characteristics Number of people in the home Year home built (MPAC) Market value of home (MPAC) Distance to expressway (GIS) Forced air heating (MPAC) Have air conditioner (central or window unit) Have central air conditioner Use air conditioning > 30 days/year Wood burning fireplace Air cleaning filter on furnace Premium air cleaning filter on furnace Dog or cat in the home Storm windows

Total N 46 46 46 46 46 35 46 46 46 46 44 44 44 46 46 46 46 46 46 45 46 46 46

Mean (standard deviation) 0.52 ±0.21 8.17 ±5.18 µg/m3 9.72 ±3.90 µg/m3 0.46 ±0.31 µg/m3 0.76 ±0.36 µg/m3 0.22 ±0.15/h 22.0 ±2.1 °C 52.0 ±7.5 % 14.6 ±6.2 °C 72.6 ±7.7% Median (quartile range) 4 (4–5) 1948 (1925–1967) $437,000 ($330,000–558,000) 1.85 km (1.3–2.6 km) Frequency (percentage) 33 (72%) 44 (96%) 39 (85%) 24 (52%) 20 (43%) 34 (76%) 16 (36%) 17 (37%) 10 (22%)


3217

The mean (±standard deviation) Finf among the 46 included homes was 0.52 ± 0.21, ranging from 0.08 to 0.88 with approximately normal distribution. Means were not significantly different based on whether homes were sampled during times when heating might occur: when the outdoor temperature was below 18 °C, the average Finf was 0.54 ± 0.22 as compared with 0.48 ± 0.21 when temperature was above 18 °C. This difference was not statistically significant (p = 0.34). The mean sulphur concentration was 0.46 ±0.31 µg/m3 indoors and 0.76 ±0.36 µg/m3 outdoors. On average, PM2.5 concentrations were slightly higher outdoors (9.72 ± 3.90 µg/m3) than indoors (8.17 ±5.18 µg/m3). The results of simple linear regression are presented in Table 2. Since the great majority of homes had air conditioners (96%), a smaller group of ―regular users of air conditioning‖ was also defined: homes that reported using their air conditioner at least 30 days/year. This group consisted of 24 homes (52%). No other information on use of heating or cooling systems was available, including whether they were actually used during sampling. Table 2. Simple linear regression of infiltration (Finf) with housing and climate characteristics predicted to influence Finf. Bold indicates regression with a p-value under 0.10. Independent Variable

N

Ln of Air exchange Absolute temperature difference between indoors and outdoors (°C) Number of people in the home Year home built Market value of home ($100,000) Distance to expressway (km) Use air conditioning >30 days/year Forced air heating (0/1) Wood burning fireplace (0/1) Air cleaning filter on furnace (0/1) Premium filter on furnace (0/1) Dog or cat in the home (0/1) Storm windows (0/1)

35 46

Regression Coefficient 0.139 −0.003

44 44 44 46 46 46 46 45 45 46 46

0.003 −0.003 −0.002 0.023 −0.201 −0.130 0.120 −0.097 0.083 0.029 0.121

0.003 0.699

Standard Error 0.043 0.007

0.24 0.00

0.933 0.011 0.849 0.381 0.001 0.078 0.056 0.198 0.218 0.656 0.113

0.037 0.001 0.012 0.026 0.056 0.072 0.061 0.074 0.067 0.066 0.075

0.00 0.14 0.00 0.02 0.23 0.07 0.08 0.04 0.04 0.00 0.06

p-value

R2

In multiple linear regression modeling, air exchange rate, use of air conditioning more than 30 days/year, and having forced air heating remained as independent predictors of Finf, predicting 38% of variability (Table 3). Without air exchange, the model was able to predict 26% of variability.


3218

Table 3. Multivariate models predicting infiltration factor (Finf) using climate and housing characteristics. Independent Variables 1

2

3

Intercept Ln of air exchange Use air conditioning > 30 d/yr Forced air heating (0/1) Intercept Use air conditioning > 30 d/yr Forced air heating (0/1) Intercept Use air conditioning > 30 d/yr Forced air heating (0/1) a

Model Regression Standard P-value Na Coefficient Error Including air exchange as potential predictor 35 0.870 0.082